Languages
Most AI detectors were built and tested in English. We test each language separately, set its own threshold on human writing in that language, and say plainly when we haven't tested a language yet.
What each status means
| Status | What we did | What you get |
|---|---|---|
| Tested | Trained on the language, tested on writing from current AI models, threshold set on human text in that language. | A document verdict and passage highlights. See the accuracy page for evaluation limits. |
| Beta | Trained on the language, but tested on less data or fewer kinds of writing. | A research verdict with limited validation. |
| Not yet | No test data we trust. | No verdict. We would rather say nothing than guess. |
Current status
| Language | Status | Notes |
|---|---|---|
| English | Tested | Largest training set. Checked against 10 AI models released in 2026. |
| Français (French) | Beta | News and web human references; current AI testing uses development data. An independent test is still needed. |
| Español (Spanish) | Beta | News and web human references; current AI testing uses development data. An independent test is still needed. |
| हिन्दी (Hindi) | Beta | Devanagari only. Joining characters inside words are kept, not stripped. We need more varied human Hindi writing before calling it tested. Hinglish (Hindi in Latin letters) is not supported yet. |
| Deutsch, Bahasa Indonesia, Русский, العربية, Български, اردو | Not yet | The current serving configuration has no calibration for these languages and returns no verdict. |
| नेपाली (Nepali) | Not yet | The current serving configuration has no Nepali calibration. |
| Other languages | Not yet | Tell us which language you need. |
Why we don't just translate
Detectors that only learned English tend to treat any unfamiliar language as suspicious, or miss AI text entirely. Machine translation also changes the very statistics detectors rely on, so translating your text to English first would make the result less reliable, not more.
Each supported language uses its own human reference scores. Performance in one language does not establish performance in another.